diff --git a/dynet_test.py b/dynet_test.py
deleted file mode 100644
index e024bd41..00000000
--- a/dynet_test.py
+++ /dev/null
@@ -1,51 +0,0 @@
-from dotenv import load_dotenv
-import dynet as dy
-from neuralogic import data
-import neuralogic.dynet as nldy
-
-load_dotenv()
-
-
-def train():
-    model = data.Mutagenesis
-
-    deserializer = nldy.NeuraLogicLayer(model.weights)
-
-    epochs = 400
-    trainer = dy.AdamTrainer(deserializer.model, alpha=0.001)
-    printouts = 10
-    seen_instances = 0
-    total_loss = 0
-
-    for iter in range(epochs):
-        if iter > 0 and iter % printouts == 0:
-            print(iter, " average loss is:", total_loss / seen_instances)
-
-        seen_instances = 0
-        total_loss = 0
-
-        dy.renew_cg(immediate_compute=False, check_validity=False)
-
-        losses = []
-
-        for sample in model.samples:
-            label = dy.scalarInput(sample.target)
-            graph_output = deserializer.build_sample(sample)
-            loss = dy.squared_distance(graph_output, label)
-            losses.append(loss)
-
-        loss = dy.esum(losses)
-        total_loss += loss.value()
-        loss.backward()
-        trainer.update()
-        seen_instances += 1
-
-    for sample in model.samples:
-        dy.renew_cg(immediate_compute=False, check_validity=False)
-
-        graph_output = deserializer.build_sample(sample)
-        label = dy.scalarInput(sample.target)
-        print(f"label: {label.value()}, output: {graph_output.value()}")
-
-
-train()
diff --git a/neuralogic/model/java_objects.py b/neuralogic/model/java_objects.py
index 61956f5c..29736ef3 100644
--- a/neuralogic/model/java_objects.py
+++ b/neuralogic/model/java_objects.py
@@ -46,9 +46,16 @@ def get_term(self, term, variable_factory):
             return self.constant_factory.construct(str(term))
         raise NotImplementedError
 
-    def get_generic_atom(self, atom_class, atom, variable_factory):
+    def get_generic_atom(self, atom_class, atom, variable_factory, new_weight):
         predicate = self.get_predicate(atom.predicate)
-        weight = self.get_weight(atom.weight, atom.is_fixed) if isinstance(atom, factories.atom.WeightedAtom) else None
+
+        weight = None
+        if isinstance(atom, factories.atom.WeightedAtom):
+            if new_weight:
+                weight = self.get_weight(atom.weight, atom.is_fixed)
+            else:
+                weight = get_field(atom.java_object, "weight")
+
         term_list = ListConverter().convert(
             [self.get_term(term, variable_factory) for term in atom.terms], get_gateway()._gateway_client
         )
@@ -126,10 +133,10 @@ def get_predicate_metadata_pair(self, predicate_metadata):
         return namespace.Pair(predicate_metadata.predicate.java_object, self.get_metadata(predicate_metadata.metadata))
 
     def get_valued_fact(self, atom, variable_factory):
-        return self.get_generic_atom(self.example_namespace.ValuedFact, atom, variable_factory)
+        return self.get_generic_atom(self.example_namespace.ValuedFact, atom, variable_factory, True)
 
     def get_atom(self, atom, variable_factory):
-        return self.get_generic_atom(self.namespace.BodyAtom, atom, variable_factory)
+        return self.get_generic_atom(self.namespace.BodyAtom, atom, variable_factory, False)
 
     def get_rule(self, rule):
         java_rule = self.namespace.WeightedRule()